暗网 YOLOv4 训练 - 它是否打印每张图像、每批次、每个细分?得到奇怪的数字

Darknet YOLOv4 training - does it print every every image, every batch, every subdivision? Getting weird numbers

当我训练 yolo v4 时,我得到了很多输出,我以后想用这些输出来绘制学习率变化。

使用 max_batches = 1543 暗网打印 74064 个输出(有损失等,我不计算 mAP 计算、网络定义等的输出)。 batch = 32,subdivisions = 16,我只有 49376 次迭代(定义为通过网络传递的单个图像)!

肯定有问题,有人知道这个二进制文件应该多久打印一次值吗?

./darknet detector train data/obj.data ../yolov4-custom.cfg yolov4.conv.137 -dont_show -map &> logs.txt & disown
[...]
v3 (iou loss, Normalizer: (iou: 0.07, cls: 1.00) Region 139 Avg (IOU: 0.000000, GIOU: 0.000000), Class: 1.000000, Obj: 0.000000, No Obj: 0.000000, .5R: 0.000000, .75R: 0.000000, count: 1, class_loss = 0.500000, iou_loss = 0.000000, total_loss = 0.500000 
v3 (iou loss, Normalizer: (iou: 0.07, cls: 1.00) Region 150 Avg (IOU: 0.394489, GIOU: 0.289983), Class: 0.985003, Obj: 0.004027, No Obj: 0.000896, .5R: 0.000000, .75R: 0.000000, count: 6, class_loss = 3.253636, iou_loss = 0.891068, total_loss = 4.144704 
v3 (iou loss, Normalizer: (iou: 0.07, cls: 1.00) Region 161 Avg (IOU: 0.612244, GIOU: 0.555591), Class: 0.897004, Obj: 0.014743, No Obj: 0.000339, .5R: 1.000000, .75R: 0.000000, count: 1, class_loss = 0.492179, iou_loss = 0.095293, total_loss = 0.587472 
[...]

似乎二进制的输出打印在 yolo_layer.c 中。 Yolo v4 有 3 个 yolo 层。 yolo层记录的batch size = 2 (global_batch_size/subdivisions)。 每批 16 次正向传球 * 1543 批次 * 每正向传球 3 次打印 = 74064

有道理。